Reinforcement learning approach to self-organization in a biological manufacturing system framework

Abstract Biological manufacturing systems (BMS) aim at dealing with complexity and uncertainty in today's manufacturing, employing biologically inspired ideas such as self-organization, learning and evolution. This study proposes a self-organizing manufacturing system where manufacturing process progresses as a result of local interaction among manufacturing elements using potential fields as an implementation of the BMS idea. The study then verifies that system's feasibility. However, it is difficult to achieve the complex global objective of a system using only the self-organization method because it uses only local information of each element to implement; achievement of global objectives often requires global system information. This study proposes a reinforcement learning approach to a self-organizing manufacturing system to achieve the global manufacturing system objectives. The proposed method is applied to the maximizing throughput problem with consideration of the machine set-up time. The results of computer simulations illustrate effectiveness of the proposed method.